Catalog, discovery, observability, governance

Data catalog, data discovery, data observability, data governance. I hear all four most days, usually used as though they are interchangeable, which they aren’t. Every time it happens I think of the line from The Princess Bride about a word not meaning what the speaker thinks it means.

I have two young kids, so the version that finally stuck for me involves toys.

Catalog is the list of every toy in the house, so my wife knows what we have and roughly where it lives.

Discovery is actually finding one so a kid can play with it.

Observability is knowing whether it still works. If it’s broken, Dad fixes it.

Governance is the rules about who plays with what and how it gets put away. Mom sets those rules. Not Dad.

Metadata is the tag on the toy telling you the name, the brand, and when it was made.

In work terms, each one answers a different question, and buying a tool for one of them settles none of the others:

Capability	Question it answers	What shows it is working
Catalog	What data exists, and what is known about it?	Current definitions, owners and lineage
Discovery	Can the right person find the right data?	Someone locates the right source without a private introduction
Observability	Can the team detect and investigate a problem?	A meaningful signal reaches someone able to act
Governance	Who decides, under which rules?	Decision rights and change processes work in practice

A comprehensive catalog with owners who left a year ago still cannot settle an argument about a contested number.

What actually worked

The failure these four words usually get bought to fix has one shape: a change ships upstream, something breaks downstream, and the people who could have objected find out from the broken dashboard.

The tempting answer is to track sources by hand, update the catalog manually, and email the company when something changes. That lasts about two weeks.

What worked instead had three parts.

Figure out who owns what downstream. This is the hard part, and no tool does it for you. You have to go and ask, model by model, and record the answer where the model lives. A catalog filled in from the schema alone has owners that were true on the day of the import.

Automate the notification. Every schema or transformation change made through dbt in the repo triggered a message in a dedicated channel, orchestrated by Airflow, naming the models that were new, updated or deleted.

Gate the merge. Downstream owners had to approve before a change could land on the main branch. In my judgment that single control stopped more breakage than the alerting did, because an alert tells you something already happened, and an approval stops it happening.

Broken pipelines and dashboards dropped substantially. I never measured how much of that came from the approval step and how much from the ownership work that landed with it.

Two years on, I would change two things. I would generate the notice from the dependency graph and send it where the affected owners already read, rather than to a channel of its own. And I would scale the review to the consequence: a breaking change to a shared contract, a renamed column, a changed grain, a redefined measure, needs the owners’ agreement before release, while a compatible change needs only the notice. Requiring every consumer to approve every change just rebuilds the central bottleneck in a different place. Tools like DataHub take the discovery half further, making dependencies visible to business and technical users alike.

If you want somewhere to start, pick one shared metric and follow it from definition to change approval to incident response. The missing owner on that path is a better starting point than a longer feature comparison.

---

What changed in 2026

I wrote the above for an audience of people: an analyst looking for a table, a data engineer trying not to break someone’s dashboard, a stakeholder wondering where a number came from. The catalog served people, and the cost of bad governance was measured in broken dashboards and annoyed colleagues.

That calculation changed when the thing querying the warehouse stopped being a person.

A person who hits an undocumented table with an ambiguous column name at least has the chance to hesitate. Not always taken, as anyone who has watched a wrong table get wired into a dashboard knows. But the pause exists, and someone occasionally asks. That hesitation was a quality control nobody designed and nobody valued, and it was doing more work than anyone credited.

A model doesn’t hesitate. Point one at ungoverned data and it will join the wrong table to the wrong table and hand back a confident, well-formatted, entirely plausible answer. Nobody catches it, because a wrong answer and a right answer look identical, and there is no broken dashboard to notice. It can be built to ask or to decline instead, but whether it does is something to test, not assume.

So the four words stopped being vocabulary hygiene and became load-bearing.

Catalog is no longer a directory people browse. It is the context the model reads to know that this table is the one finance uses and that one is a deprecated copy.

Discovery matters less as a human workflow and more as a retrieval problem. The question is no longer whether a person can find the right dataset. It is whether the system surfaces it, and not the four things next to it.

Observability has to extend to the answers, not just the pipelines. Knowing your job ran is table stakes. Knowing whether what the system told someone was correct is a different discipline, and most data teams haven’t built it yet.

Governance turned into the thing that decides whether any of it works. Not the compliance chapter at the end. The prerequisite.

The practical shape of that is to invest at the unglamorous end first: owners and settled definitions in the catalog, written so a system can read them, and a regular check on what the system says back. I made the same argument about BI as a whole in What BI means, revisited.

The metaphor still holds, unfortunately. The toys are cataloged and the rules are on the wall. The difference is that something in the house now plays with all of them at once, very fast, and never asks whether it’s allowed.